重疊轉換用在聲學回音消除器的研究

Abstract

在遠距會議系統(teleconference system)或免手式電話(hands-free telephone)中，由擴音器放出來的語音信號會被傳送回發話者而形成所謂 的遠端回音。此種回音可由聲學回音消除器予以消除。這□我們提出一個 用DWT為快速演算法中建構方塊的重疊轉換所形成的聲學回音消除器，並 討論了它與其他方法在性能與複雜度上的比較。使用此一提出的結構，會 有比其在Malvar所著一書中之原始結構有更好之 ERLE及追蹤(tracking) 能力。它比該原始結構有較少的計算上的負擔(指乘法數)，當然也比正交 轉換LMS的架構需要更少的乘法量。此外，我們也做了以語音為輸入信號 的性能測試。最後，我們展示此處所提出的架構有較好的性能及效率。 The speech broadcasts from the loudspeaker of the tele- conference system or the hands-free telephone can be transmitted back to the talker as a far-end talker-echo .The far-end talker- echo can be eliminated by the echo cancellation technique using an acoustic echo canceller. An acoustic echo canceller based on the lapped transform using DWT as the building block for the fast algorithm is proposed . We compare the proposed structure with other methods in terms of performance and complexity . With the proposed structure, we can get better ERLE and tracking capability than the original lapped transform structure presented by Malvar [16] . It has less computational load than the latter and much less computational load than the orthogonal transform LMS structure . Performance tested by using a speech input is also completed .